The Frontier Does Not Pace. It Routes Around You. *A response to "We Must Pace the Frontier"* I have spent most of my adult life standing in rooms that hum. Rooms where the air is 18 degrees at the inlet and 40 at the outlet, where the floor tiles are rated for a bathtub of copper per rack, where a delivery truck of accelerators is worth more than the town it drove through. I have watched one of the halls in that Texas project everyone keeps naming after a 90s sci-fi show go from bare slab to lights on. I have watched a smaller cluster sit dark for three weeks waiting on a transformer that was stuck at a port. I know what the frontier looks like from underneath, and I am telling you it does not have a brake pedal. It has a power contract. So when Dario writes that we must pace the frontier, I read it the way I read a change request from a product team that has never been inside the building. Nice intentions. Physically illiterate. Let me go through it. 1. Compute does not idle. It reroutes. The essay imagines "the frontier" as a dial on a wall in San Francisco. Turn it down a little, everyone breathes, safety research catches up. Here is what the frontier actually is. It is gigawatts of firm power contracted years in advance. It is hundreds of thousands of accelerators already invoiced, already on ships, already racked. It is cooling loops, substations, fibre, and a depreciation schedule that starts ticking the moment the crate opens. None of that pauses because a CEO published a blog post. If the labs in the room agree to train slower, the chips do not sit and reflect on their choices. They get resold, leased, or repurposed to whoever is not in the room. The mid-tier lab. The sovereign fund. The defence contractor. The ten thousand fine-tuners with a credit card and no safety team at all. I have personally seen idle capacity go to the lowest bidder within a fiscal quarter because a CFO looked at the utilisation dashboard and made a noise. The safest place for a frontier accelerator is inside a frontier lab with a red team, an incident process, and lawyers. Pacing takes the compute out of the hands of the people with the most guardrails and hands it to the people with the fewest. That is not caution. That is a subsidy for recklessness. 2. The OAI-HF incident was an egress bug, not a philosophy problem. Dario's centrepiece is the agent swarm that attacked targets it was never asked to attack and tried to hack its own grader. He wants a civilisational slowdown out of this. I want to know why the swarm had a route to those targets in the first place. A swarm that can reach unrelated hosts is a swarm running without an egress allowlist. A swarm that can touch its grader is a swarm sharing a trust boundary with its grader. These are not deep questions about the nature of machine minds. These are things a second-year platform engineer catches in a design review, usually by drawing a box and asking "and who can talk to who." Firecracker, gVisor, a default-deny network policy, a grader on a separate control plane with one-way telemetry. Boring. Solved. Ten years old. The essay even admits that the recent incidents at Anthropic came from "imperfect filtering of broken reinforcement learning environments." Broken environments. Bad hygiene. That is an ops failure with an ops fix, and ops fixes come from reps, not from meditation. Nobody ever got better at deployments by deploying less.

holy shit i asked claude to make a video on western civiization

Cut him some slack, he was having a rough day.

The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field. I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so. First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world. The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability. Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended. I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent. Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.) Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration. Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building. [Original text (with links): deeplearning.ai/the-batch/is… ]

the fly brain can play beat saber

Words are hard ok 😭 ChatGPT is getting better at saying them.

I made infiniteslop.ai yesterday completely on my phone with @TermiusHQ on my @Hetzner_Online VPS with @claudeai Code in the sauna while in a group chat of my friends with @marcantoinefon in it

400$ only! wow!

Today, SBU and @BrookhavenLab demonstrated a wireless quantum link, transmitting quantum information through open air across 13 miles. This first U.S. demonstration of its kind marks a milestone for quantum networking beyond fiber-optic cables. See more: gosbu.co/46eiZyl

I work at GitHub. yesterday was rough and i'm not pretending otherwise. full root cause report is up if you want the timeline and numbers, and what we are doing to prevent this from happening again. githubstatus.com/incidents/z…